The best subset selection problem solved with mixed integer optimization
Specialeforsvar ved Trine Sevel Lollike
Titel: The Best Subset Selection Problem Solved With Mixed Integer Optimization
Abstract: We consider the classical best subset selection problem solved with a modern mixed integer optimization approach, used for variable selection in linear regression. This thesis presents the theory behind reformulating the best subset selection problem into a mixed integer optimization problem and considers the theoretical and algorithmic advantages as well as disadvantages of this reformulation. The main contribution is an extended comparison, of the modern mixed integer optimization approach to the existing regression methods; lasso and forward selection, in a realistic setup.
Vejleder: Niels Richard Hansen
Censor: Alexander Sokol, Nordea